动态场景多目实时拼接系统的设计与实现
Design and implementation of a real-time multi-camera stitching system for dynamic scenes
卓越 1刘伟1
作者信息
- 1. 北京邮电大学计算机学院，北京 100876
- 折叠
摘要
多相机视频拼接中动态目标越过接缝往往会形成较大的重影，且拼接系统实时性一般。本文设计并实现了一个动态场景多目实时拼接系统。该系统采取动静分离的思想，在预处理阶段，事先离线计算静态背景的重映射矩阵与接缝；在线处理时结合多线程异步调度机制，使用实例分割模型提取流媒体中的行人、车等动态前景，之后融合重映射、色彩补偿等实现图像的无缝拼接处理。实验结果表明，4路并行视频流可以通过该系统合并为实时全景画面，同时消除接缝动态重影，全栈处理帧率基本能稳定在40～60帧之间，较好地解决了多目视频拼接的计算时延问题。
Abstract
In multi-camera video stitching, dynamic targets crossing the seam often cause significant ghosting artifacts, and the real-time performance of stitching systems is generally limited. This paper designs and implements a real-time multi-camera stitching system for dynamic scenes. The system adopts a separation-of-dynamic-and-static approach: in the preprocessing stage, the remapping matrices and seams for the static background are computed offline; during online processing, with a multi-threaded asynchronous scheduling mechanism, an instance segmentation model is used to extract dynamic foreground objects such as pedestrians and vehicles from the streaming media. Subsequently, remapping, color compensation, and other techniques are integrated to achieve seamless image stitching. Experimental results show that four parallel video streams can be merged into a real-time panoramic view using the proposed system, while eliminating dynamic ghosting at the seams. The end-to-end processing frame rate can stably reach 40-60fps, effectively addressing the computation latency issues in multi-view video stitching.关键词
计算机应用/多目视频拼接系统/动静分离方法/动态场景/多线程调度Key words
computer application/multi-camera video stitching system/static-dynamic separation method/dynamic scenes/multi-thread scheduling引用本文复制引用
卓越,刘伟.动态场景多目实时拼接系统的设计与实现[EB/OL].(2026-06-22)[2026-06-24].http://www.paper.edu.cn/releasepaper/content/202606-60.学科分类
计算技术、计算机技术
